LIFE Med Hiss: an innovative cohort design for public health

(1)

Method

Article

LIFE

Med

Hiss:

An

innovative

cohort

design

for

public

health

Martina

Gandini

a,b,

*

,

Cecilia

Scarinzi

b

,

Stefano

Bande

c

,

Giovanna

Berti

b

,

Luisella

Ciancarella

d

,

Giuseppe

Costa

b,e

,

Moreno

Demaria

b

,

Stefania

Ghigo

c

,

Chiara

Marinacci

f

,

Antonio

Piersanti

d

,

Gabriella

Sebastiani

g

,

Ennio

Cadum

b a

UniversityofTorino,DepartmentofClinicalandBiologicalScience,AOUSanLuigiGonzaga,RegioneGonzole 10,10043,Orbassano,Turin,Italy

b_{Environmental}_{Epidemiological}_Unit,_Regional_{Environmental}_Protection_Agency,_Piedmont_Region,_Via_Pio

VII9,10135,Turin,Italy

c

AirQualityUnit,RegionalEnvironmentalProtectionAgency,Piedmont,ViaPioVII9,10135,Turin,Italy

d

LaboratoryofAtmosphericPollution,ENEA-BolognaResearchCenter,ViaMartiridiMonteSole4,40129, Bologna,Italy

e

RegionalEpidemiologyUnit,ASLTO3PiedmontRegion,ViaSabaudia164,10095,Grugliasco,Italy

f_Department_of_{Epidemiology,}_Lazio_Regional_Health_Service,_Rome,_Italy g_National_Institue_of_Statistics,_Rome,_Italy

ABSTRACT

TheaimofMEDHISSmethodologywastotesttheeffectivenessofalow-costapproachtostudylong-termeffects

of air pollution, applicable in all European countries. This approach is potentially exportable to other

environmentalissueswhereacohortrepresentativeofthecountrypopulationisneeded.

ThecohortisderivedfromtheNationalHealthInterviewSurvey,compulsoryinEuropeancountries,

whichhasinformationonindividuallifestylefactors.InLifeMedHissapproach,subjectsrecruitedhave

beenlinkedatindividuallevelwithhealthdataandhavebeenthenfollowed-upformortalityandhospital

admissionsoutcomes.Exposurevaluesofairpollution(PM2.5andNO2)havebeenassignedusingnational

dispersion models, enhanced by the information derived from monitoring station with data fusion

techniques,and thenupscaledat municipalitylevel(highestlevelof detailachievable fortheItalian

Survey).Results for mortality have been used to test theeffectiveness of this methodology andare

encouragingifcomparedwithEuropeanones.

Theadvantagesofthistechniquearesummarizedbelow:

*Corresponding authorat:EnvironmentalEpidemiologicalUnit,RegionalEnvironmentalProtectionAgency,Piedmont Region,ViaPioVII9,10135,Turin,Italy.

E-mailaddresses:martina.gandini@unito.it(M.Gandini),cecilia.scarinzi@arpa.piemonte.it(C.Scarinzi),

stefano.bande@arpa.piemonte.it(S.Bande),giovanna.berti@arpa.piemonte.it(G.Berti),luisella.ciancarella@enea.it

(L.Ciancarella),giuseppe.costa@epi.piemonte.it(G.Costa),moreno.demaria@arpa.piemonte.it(M.Demaria),

stefania.ghigo@arpa.piemonte.it(S.Ghigo),c.marinacci@deplazio.it(C.Marinacci),antonio.piersanti@enea.it(A.Piersanti),

gabriella.sebastiani@istat.it(G.Sebastiani),ennio.cadum@ats-pavia.it(E.Cadum).

https://doi.org/10.1016/j.mex.2018.12.007

ContentslistsavailableatScienceDirect

MethodsX

(2)

ItusesacohortalreadyavailableandcompulsoryinEuropeancountries

Itusesairqualitymodellingdata,availableformostofthecountries

Itpermitstoimplementversatileenvironmentalsurveillancesystems

creativecommons.org/licenses/by/4.0/).

ARTICLE INFO

Keywords:Airpollution,Epidemiologicalsurveillance,Longtermstudies,NationalHealthInterviewSurveys,Exposure assessment

Articlehistory:Received11October2018;Accepted15December2018;Availableonline18December2018

SpeciﬁcationsTable

SubjectArea EnvironmentalScience

Morespeciﬁcsubjectarea: SystemtomeasureadversehealtheffectofairpollutionusingacohortderivedfromNational HealthInterviewSurvey

Methodname: LIFEMedHissprojectapproach

Nameandreferenceoforiginal method

NA

Resourceavailability NHIS(https://www.cdc.gov/nchs/nhis/index.htm) MINNI(http://www.minni.org/?set_language=en) BRACE(http://www.brace.sinanet.apat.it/)

EuropeanAirQualityDatasetAirbase(http://www.eea.europa.eu/)

CorineLandCover(https://land.copernicus.eu/pan-european/corine-land-cover) R(https://cran.r-project.org/bin/windows/base/)

SAS(https://www.sas.com/it_it/home.html)

Methoddetails

LIFE MED HISS project aimed at consolidating the knowledge base for the development, assessment,monitoringandevaluationofenvironmentalpolicyandlegislation,bysettingupa low-costEuropeansurveillancesystemoflong-termeffectsofairpollutionbasedoncohortsrecruited usingNationalHealthInterviewSurvey(NHIS)dataalreadyavailable.

A large number of epidemiological studies have linked long-term concentrations of air pollutiontomortality,butrelativelyfewstudiesfocusedonthelongtermeffectsofparticulate (PM10 and PM2.5)or NO2 onhospital admissions. As mortality risk factor, in2015, ambient

PM2.5wasclassiﬁedas theﬁfth-ranking.Mostof thecohortstudieswere carriedoutinUSA, whereboththecharacteristicsandsusceptibilityofthepopulationexposedandthecomposition ofairpollutionmixturecanbedifferentfromEuropeancontext.Severalcohortsareavailablein Europe,asthenationalEnglishcohort(812,063patientsaged40–89yearsasreportedby[1]),the DanishDiet,Cancer,andHealth(DCH)cohort(57,053peoplefromCopenhagenorAarhusaged50– 65years,asreportedby[2])ortheLondoncohort,whichstudiedalsocombinedeffectsofnoise andairpollution[3].

The cohort werecruited is based onthe National(Italian)Health Interview Survey, which is compulsoryasin allEuropean countriesand hasinformation onindividuallifestylefactors.This cohorthasbeenfollowed-upformortalityandmorbidity.Forexposureassessmentitwaspossibleto assigntoeachindividualtheexposurevaluetoairpollution(PM2.5andNO2),derivedfromnational dispersionmodels,improvedwithmonitoringstationusingdatafusiontechniques,andthenupscaled atmunicipalitylevel.

(3)

Healthdata

Thefollow-upisbasedonthesampleincludedinthe1999–2000NHIS,carriedoutbytheNational InstituteofStatistics(ISTAT),linkedtomortalityandhospitalrecords[4].TheoriginalNHISsample consistsof140,011individualsbelongingto52,232sampledfamiliesresidentin1449municipalities inthewholenationalterritory(Fig.1).TheNHIScontains,besideothers,individualinformationon age,gender,occupationalandeducationalstatus,maritalstatus,residence,bodymassindex(BMI), smokinghabitandphysicalactivity.Informationonalcoholwasnotavailableinthesurvey,while informationondietwaslimitedtoparticularconditions(e.g.macrobioticdiet)andthereforecouldnot beusedtoassesslifestyles.InFig.1thereisamapwiththenumberofsubjectssampledineach municipality.

(4)

To perform record linkage, the personal data from the paper format of family status of the interviewedpeoplemadeavailablebyISTATwererecorded.Attheendoftheregistrationandcontrol operations,anarchivewasobtainedwiththepersonaldataof139,991individuals.Thisarchivewas linkedtothecohortoftheinterviewees(140,011).Usingtheinformationfrombotharchives,theFiscal Code necessary for the record linkage with the other sources was calculated. Of the 140,011 respondents,128,818individualswerelinkablewithfullorpartialfiscalcode,representing92%ofthe HISsample.ArecordlinkagehasbeencarriedonformortalityandhospitaladmissionswiththeISTAT nationalarchiveofthecausesofdeath.Thetaxcodewasusedasrecordlinkagekey.Whenmissingin the death certificate, tax code was imputed based on the reported personal information. The codificationofcausesofdeathandhospitalizationhasbeenperformedusinganautomaticprocedure: therejectsthatcannotbeautomaticallyprocessed(about20%ofthetotal)aremanuallycoded.The underlyingcauseisselectedaccordingtointernationalrulesfromtheWorldHealthOrganization, reportedintheInternationalClassificationofDiseases,ninthrevision(ICD-9),forallcasesupuntil 2002andinInternationalClassificationofDiseases,tenthrevision(ICD-10),forallcasesfrom2003 onwards. Only for interviewed individuals withcomplete fiscal code (126,601) a recordlinkage between 1999–2000 NHIS and hospital dischargesarchives of the Ministry of Health has been performed.Hospitaladmissionswithfiscalcodewereavailableonlystartingfromyear2001.

Recordlinkagehasbeenthemostcriticalpointinthesettingupoftheproject.Twoparticipating countries(SpainandFrance)couldnotapplythismethodologyduetoprivacylegislationandhadto carryonanothermethodology[5].Sloveniainstead,tookadvantageofthisprojecttoimproveprivacy regulation.BytheendoftheprojecttheyhavebeenabletoenrolaSloveniancohortandthiskindof methodologywillbeappliedwhenanadequatenumberoffollow-upyearswillbereached.

Toevaluatethecompletenessoftherecordlinkagebetweenthecohortandthedatakeptinthe deathrecords_’nationalarchive,thenumberofdeathsobservedinthesamplewascomparedwiththe expectednumber.ThenumberofexpecteddeathshasbeenobtainedbymultiplyingtheItalianage andsex-speciﬁcmortalityratesobservedineachfollow-upyearwiththeactualnumberofindividuals ofthecohortatrisk ofdeathforeachyear,forthespeciﬁcsexandage(person-years).Asimilar procedure has beenapplied toevaluate record linkagewith hospital admissions. Bothchecking operationssuccededinpositiveresults.Moreover,descriptivestatisticsontheoriginalsampleand that one resulting from the follow-up have been computed to avoid possible distortion in the representativenessoftheItalianpopulation.

Weinvestigatedsixcausesofmortality:non-accidental(ICD-9code001–799,ICD-10code A00-R99),deathforCardiovascularDiseases(CVD)(ICD-9code390–459,ICD-10codeI00-I99),diseasesof therespiratorysystem(ICD-9code460_–519,ICD-10codeJ00-J99),neoplasmexcludinglungcancer (ICD-9code140–239except162,ICD-10codeC00-D48exceptC33–C34),lungcancer(ICD-9code162, ICD-10codeC33-C34)anddiseasesofthenervoussystem(ICD-9code320–359,ICD-10codeG00– G99).Wedecidedtoinvestigateneoplasmexcludinglungcancertoconsiderseparatelyacauseof deathwhichisstronglyassociatedwithairpollutionaccordingtoliterature.

Weexcludedfromtheanalysesnon-accidentalcausesandrepeatedhospitalizationsforthesame subject(withtheexceptionsofmyocardialinfarction,anginapectorisandlowrespiratorytractinfection (LRTI)).Foreachcause,weconsideredonlythemaincauseofhospitalizationoutofsixavailable,and,for eachsubject,onlythefirsthospitaladmission,excludingallsubsequenthospitalizationforthesame causeandforthesamesubject.Thiscriterionpermitstohaveagoodapproximationtoanincidence measure.Anexceptiontothisrulewasdonefordiabetesandchronicobstructivepulmonarydisease (COPD),whichweresearchedamongallsixcodes.Asecondtypeofexceptionwasdoneformyocardial infarction,anginapectorisandLRTI:forthefirsttwocauses,asecondeventthatoccurredafter28daysor morefromthefirstepisodewasconsideredasanewevent[16],whereasforLRTIthetimeelapsedfrom thefirstepisodetoconsideritasaneweventhadtobe90daysormore.

Sincethetopicwasrelativelynewconsideringhospitalizationsasoutcome,weselectedawider numberofcauses. Firstly,thefollowingmaingroupof causeshavebeenanalyzed:disease ofthe circulatory system(ICD-9code390–459),heartdiseases(ICD-9code390–429),cerebrovasculardiseases (ICD-9code430–438),diseasesoftherespiratorysystem(ICD-9code460–519),allneoplasmbutlung cancer(ICD-9code140–239,withtheexceptionof162),mental,behavioralandneurodevelopmental disorders(ICD-9code290–319),diseasesofthenervoussystem(ICD-9code320–359).Secondly,we

(5)

investigatedthefollowingspeciﬁccauses:lungcancer(ICD-9code162),bladder cancer(ICD-9code188), kidneycancer (ICD-9code189),diabetes (ICD-9code250),Parkinson’s disease(ICD-9 code332), Alzheimer’sdisease(ICD-9code331),myocardialinfarction(ICD-9code410),anginapectoris (ICD-9code413)atherosclerosis(ICD-9code440),LRTI(ICD-9code466,480–487),COPD(ICD-9code490– 492,494,496),asthma(ICD-9code493),andmiscarriage(ICD-9code634).

InFig.2thereisagraphicalrepresentationofthefollow-upscheme.Anindividualwhichtookpart totheNHISmayexperienceoneoftheoutcomesofinterestduringthefollow-up(untilyear2008), afterthefollow-up,ormaynothaveexperiencedtheeventtothepresentday.

DuetothenatureoftheLIFEMEDHISSproject,variablesincludedinthemodelascovariatesare harmonized, accordingto the standard developed within theEuropean EUROTHINEproject [6], ensuringahighdegreeofcomparabilitywithotherEuropeansurveys,alsoconsideringlevelschosen forcategoricalvariables. Covariatesincludedin themodelare: gender,educationallevel, marital status,occupationalstatus,smokinghabit,physicalactivity,BMI,typeofmunicipality.

Exposuretoairpollution

DuetothenatureoftheNHIS,whichhasindividualssampledbothinruralandurbanareasofthe wholenationalterritory,thebestchoicetoassignexposuredatamaybetheuseofmodellingdata. Sinceinhealthdataitwaspossibletoknowonlyresidence’smunicipality,withnoinformationon addressorzipcode,acomparablenationalcoveragehasbeenneededforairqualitydata.

TheItalianexposuremodelconsidersdatacomingfromthemodelingsystemMINNI(National IntegratedModellingsystemforInternationalNegotiationonatmosphericpollution),developedand validatedbyENEAandtheItalianMinistryoftheEnvironmentLandandSea[7,8].MINNIincludesan atmosphericchemicaltransportmodel(CTM)fedbyemissiondata(nationalemissioninventoryof anthropogenicsources,biogenicVOCs,sea-salt,naturaldust)andmeteorologicalprognosticﬁelds. MINNIoutputsforPM2.5andNO2wereprovidedforyears1999,2003,2005and2007ataspatial

resolutionrepresentativeofurbanbackgroundpollutionlevels(44km2)athourlyresolution([7]b). TheCTMmodelincludedintheMINNImodellingsystemiscalledFARM[13,15]andtheversionusedin thestudyisthe3.1.12.

Toreducemodeluncertainty,adatafusiontechniqueisusedtocombineMINNIconcentration ﬁeldsandpollutantobservations.Thus,itispossibletoprovidea morerealisticrepresentationof pollutantspatialdistribution.Duetothenatureofthehealthdata,onlybackgroundstationshavebeen consideredtointegrateCTMmodels.Observeddataintroducedinthedispersionmodeloutputsare retrievedfromtheItalianregionalenvironmental agenciesBRACEdatabase(Thenationalsystem: http://www.brace.sinanet.apat.it/)andfromtheEuropeanAirQualitydatasetAirBase(TheEuropean air quality database: http://www.eea.europa.eu/

(6)

quality-database-7).Onlymonitoringstationswithlessthan30%missingdatahavebeenconsidered intheassimilationprocedure.

Therefore,aKrigingwithExternalDrift(KED[9],and[10])procedurehasbeenusedtoaccountfor theobserveddataintoMINNIﬁelds.Speciﬁcally,thekrigingisappliedontheobserveddataandthe externaldriftisconstitutedbytheMINNImodeloutput.Observationswereinterpretedasrealizations ofaGaussianspatialprocessY(s)atspatiallocations,inthedomainS.Y(s)hasthefollowingstructure:

YðsÞ¼

m

ðsÞþwðsÞþ

e

ðsÞ;

where in thetrendcomponent

m

(s) theMINNI modeloutputis introduced,w(s) is a stationary Gaussian random processand

e

(s) is theerror term.Regardingobserveddata introducedin the dispersionmodeloutputs,weretrieveddatafromotherItalianregionalenvironmentalagenciesoutof BRACE database (national system: http://www.brace.sinanet.apat.it/). Only background stations, whosespatialrepresentativenessisconsistentwiththeMINNIresolution,andonlymonitoringsites withmorethan80%ofdatahavebeenconsideredandaveragedtoobtainannualobservedvalues[11]. PM2.5andNO2arethemostimpactingpollutantsontheItalianterritory,intermsoflong-term

effectonhumanhealth.Also,SO2andCOhavehistoricallybeenofhighimpact,buttheirairborne

levelshavebeendrasticallyreducedbeforeyear2000,thereforetheirimpactisverylowinthetime spananalyzedinthisstudy.BlackcarbonwasincludedinofﬁcialEUemissioninventoriesin2017,so nonationalemissiondatawereavailableatthetimeofthisstudy.

SincethehealthdatafromtheNHISareaggregatedatthemunicipallevel,theexposureassessment must rely on the same spatial scale. Therefore, following this data fusion approach, theItalian exposureassessmentwascarriedoutbyup-scalingthegriddeddataatthemunicipalitylevel[10]. Theseannualexposuremapswereobtainedoverlying griddedconcentrationdatatomunicipality boundariesandthenusingaweightedaverage,wherebuilt-upareapercentagesweretheweights. Built-upareainformationhasbeencollectedfromCorineLandCoverdata[12]usinglanduseclasses belongingtoArtiﬁcialSurfacescategories.Inthisway,forbothpollutantsameanannualvaluehas beencalculatedforallItalianmunicipalities(Fig.1).

Thechangeofsupportproblemwassolvedbymeansofanupscalingtechnique,sincehealthdata areatmunicipalitylevel,ascale(blocks)largerthanthespatialscaleofmodelcalculation,which consistsofpointsregularlyspacedonagrid.

Sinceourspatialdomainwasdiscretizedbymeansofgridpoints(orcells),weidentiﬁedtheni pointsbelongingtotheithmunicipalityemployinggeographicinformationsystem(GIS)software. Therefore,weimplementedaweightedblockaveragingprocedurethatcanbewrittenas

Ci¼

Xni

p¼1

g

ipyp

whereCidenotestheconcentrationmeanontheithmunicipality,yprepresentstheconcentration

valueofthepthcelland

g

ipistheweightforthepthgridpointoftheithmunicipality.

Theweightwaschosenas

g

ip¼

Aip

Ai

where Aip denotes the municipality built-up area belonging to the pth cell and Ai the whole

municipalitybuilt-uparea.Thiskindofweightwaschosensincebuilt-upareasareassociatedwith anthropicactivitiesandsorepresentpartofthemunicipalitywheremostpeoplespendtheirtime, includingresidentialandworkingareas.

InFig.3thereinanexampleofoneyearofexposureassessment(2005),consideringPM2.5as pollutant.

Statisticalanalyses

For each outcome analysed, the individual effect of long-term exposure to air pollution on mortality and hospitalization is modelled using the Cox proportional hazard model, with air

(7)

pollutionlevelsandageclassastime-varyingvariables,whileadjustingforothervariables.Robust varianceestimatesareproducedtotakeintoaccountthetwo-stagesamplingstrategyusedbyISTAT (Istituto Nazionale di Statistica) in the survey (sampling municipalities at the ﬁrst stage and householdswithineachmunicipalityatthesecondstage,andenrollingalltheindividualsofthe sampledfamilies).

Weestimatehazardratios(HRs)adjustedforgender,educationallevel(classifiedaccording to the international standard classification of education (ISCED) and grouped into the four classes:uptoprimary,correspondingtoISCED0–1,lowersecondary,ISCED2and3C,uppersecondary, ISCED3Aand3Bandpostsecondary,ISCED4–6),maritalstatus(livingwithpartner/notlivingwith partner), occupational status (employed/not employed), smoking habit(current smoker, former smoker,never smoker), physical activity(4 different levelscombining information on type and frequencyofactivity) and BMI.The individualcharacteristicsareknownonlyat thetime of the interview.Foreachsubjectwedefinetheresidencebasedontypeofmunicipalityaccordingtothe populationsize:metropolitanareas(municipalitieswithmorethan250,000inhabitants),urbanareas (between 20,000 and 250,000 inhabitants) and rural areas (municipalities with less than 20,000inhabitants).

Resultsformortalityhavebeenusedtotesttheeffectivenessofthismethodology,comparingHRs withthoseobtainedintheliterature.

TheideaofLIFEMEDHISSprojectistoobtainthemaximuminformationfromsourcesalready available. Since we had four years of air pollution exposure (from 1999 to 2007), which were subsequenttotheenrollmentofthecohort,wedidnot_ﬁndappropriatetouseairpollutionas time-dependentvariableusing only themost recent yearof exposure, as donein other studies[14]. Therefore,wedividedthefollow-upperiodin4risksetsandcomputedtheannualexposurevalueof eachrisksetasthemeanexposureofalltheprecedingyearsforwhichthedatawasavailable,as follows:

1999–2002(exposurein1999)

2003–2004(meanexposureof1999and2003) 2005–2006(meanexposureof1999,2003and2005) 2007–2008(meanexposureof1999,2003,2005and2007)

Sensitivityanalyseshavebeendoneusingonlyoneyearofexposure(2005). Fig.3. Exampleofoneyearexposureassessment–PM2.5year2005.

(8)

Theproportional hazardassumptionwas testedfor alltheﬁxed predictors and stratiﬁedCox modelswereappliedforpredictorsthatdidnotmeettheassumption.

Foreachoutcome,weevaluatedpotentialeffectmodiﬁcationbyaddingintothemodeltheproper interactiontermofexposurebymodiﬁer.Weusedthelikelihoodratiotesttocomparethemodels withandwithoutinteractionterms.

DuetotheheterogeneousspatialcoverageofmonitoringstationsinNorthernandCentralItaly (densespatialcoverage)andSouthernItaly(sparsespatialcoverage),sensitivityanalyseshavebeen donealsorestrictedtothe_ﬁrstdomain.

Allstatisticalanalyses wereconductedusing SAS,version 9.4(SAS Institute,Cary, NC) andR, version3.3.3.

Advantagesanddisadvantagesoftheapplicationofthemethod

MethoddevelopedinMED-HISSprojectaimedatcontributingtotheupdatinganddevelopmentof EuropeanUnionenvironmentalpolicyandlegislation,intermofadversehealtheffectofairpollution. The current understanding of theassociation between long-term exposure to air pollution and adversehealtheffectisbasedoncohortstudiesfromUSA,Canada,Japan,China.Fewstudieshavebeen conductedsofarinEurope,withrestrictionsonageofcohortrecruited,pollutantstudied,andwith poorgeographicalvariability.TheaimofMED-HISSmethodologyistosetupaEuropeansurveillance systembasedonretrospectivecohortsrecruitedusingNationalHealthInterviewSurvey(NHIS)data, which are compulsory in European Union, and therefore already available. Cohorts have been followed-upformortalityandhospitaladmissions.Inthisstudyweusedthecohorttostudylongterm effectsofairpollution.Toeachsubjectwillbeassignedtheexposuretoairpollution(PM2.5andNO2), derivedfromnationaldispersionmodels.

Themajoradvantagesare:

-itisalow-costapproachifcomparedwiththeotherstudiesinthisfield,whichrecruitedpeople specifically forthestudy. Tosetupa cohortimpliesthecoststoselectthesampleand fillout questionnaire,whichwithMEDHISSmethodologyarecoveredbytheinstitutionswhichcollect dataforNHIS

-inthiswaythecohortsarerepresentativeofallpopulations(urban,ruralandmetropolitanareas), spreadoverthewholenationalterritoryandwithindividualchacteristics(e.g.education,working conditions,lifestyle)representativeofthewholepopulation

-itcanbeusedforotherenvironmentalissues,notnecessarilyonlytostudyhealtheffectofair pollution

-variablesconcerningindividualcharacteristicscanbeeasilystandardizedacrosscountries,asthey comefromEuropeanNHIS.

Thedisadvantagesare:

-itcanberestrictedduetoprivacyregulation.Forexample,inSpainandinSlovenia,itwasnot possibletoapplythismethodwithintimetableofLIFEMEDHISSproject.InSpain,theidentiﬁcative keysofthesubjectsinvolvedintheNHIShavebeendestroyed,sothatthelinkagehasnotbeen possible.However,thisproblemhasbeenbroughttotheattentionofthelegislation,succeededin thelinkagewithhealthoutcomeswiththelastSlovenianNHIS(buttostudyairpollutioneffectthe yearsoffollow-uphavenotbeenenough).

-variablesconcerningindividualcharacteristicsarecollectedonlyatbaselineandcannotbeupdated (particularlyforthosewhohadnotexperiencedtheevents)

Additionalinformation

WediscouragetheapplicationofthismethodologytobothPM10andPM2.5.Inourexperience,dueto theattributionofameanannualvaluetoa wholemunicipality,thetwocomponentsofparticulatematter

(9)

arehighlycorrelated.Instead,ifitispossibletogoindepthwithinmunicipalitywithinformationof subjects’address,thisissueshouldnotbeaproblem.Moreover,particularlyinyears1999and2003, assimilationprocedureforPM2.5hasbeendonederivingthemonitoringstationsdatafromPM10.

Thesensitivityanalysiswithonly2005annualmeanasexposureisduetoalowspatialcoverageof monitoringstationsfortheyear1999.

Althoughit hasbeenpossibletorecruitthecohort andanextensionof thefollow-up willbe available,wehadfewlimitationsaswell:

Theimpossibilitytogethomeaddressorzipcode,whichwouldhavebeenensuredahigherdegree ofaccuracyinexposureassessment

The impossibility tohave information onresidential history (we onlyhad theinformation at baseline)

Thenumberofyearsoffollow-updependsonthetopicandnotonthemethodology.Tostudy long-termeffectofairpollutionweneededseveralyearsfromthetimeoftheinterviewtoevaluatethe healtheffectofairpollutionontheoutcomesunderinvestigation.

Acknowledgements

ThisresearchwassupportedbytheEULifeProgramme(grantnumberLIFE12ENV/IT/000834).The authorsdeclaretheyhavenoactualorpotentialcompeting_ﬁnancialinterests.Theauthorswouldlike tothanktheanonymousreviewersfortheirhelpfulandconstructivecommentsthatcontributedto improvingtheﬁnalversionofthepaper.TheauthorswishtothankNinoKuenzlifromSwissTropical andPublicHealthInstitute,XavierQuerolfromInstituteforEnvironmental.AssessmentandWater Research IDAEA-CSIC (Spanish Research Council) and Claudia Galassi from Unit of Cancer Epidemiology“Cittàdellasaluteedellascienza”Hospital(UniversityofTurin)andCenterforCancer PreventionPiemontefortheirassistanceduringtheprojectandtheirvaluablesuggestions. References

[1]R.W.Atkinson,I.M.Carey,A.J.Kent,T.P.vanStaa,H.R.Anderson,D.G.Cook,Long-termexposuretooutdoorairpollutionand theincidenceofchronicobstructivepulmonarydiseaseinanationalEnglishcohort,Occup.Environ.Med.72(January(1)) (2015)42–48.

[2]Z.J.Andersen,K.Bønnelykke,M.Hvidberg,S.S.Jensen,M.Ketzel,S.Loft,etal.,Long-termexposuretoairpollutionand asthmahospitalisationsinolderadults:acohortstudy,Thorax67(January(1))(2012)6–11,doi:http://dx.doi.org/10.1136/ thoraxjnl-2011-200711Epub2011Sep2.

[3]J.I.Halonen,A.L.Hansell,J.Gulliver,D.Morley,M.Blangiardo,D.Fecht,etal.,Roadtrafﬁcnoiseisassociatedwithincreased cardiovascularmorbidityandmortalityandall-causemortalityinLondon,Eur.HeartJ.36(October(39))(2015)2653–2661. [4]C.Marinacci,F.Grippo,M.Pappagallo,G.Sebastiani,M.Demaria,P.Vittori,N.Caranci,G.Costa,Socialinequalitiesintotal andcause-speciﬁcmortalityofasampleoftheItalianpopulation,from1999to2007,Eur.J.Publ.Health23(2013)582–587. [5]C.deKeijzer,D.Agis,A.Ambrós,G.Arévalo,J.M.Baldasano,S.Bande,etal.,Theassociationofairpollutionandgreenness withmortalityandlifeexpectancyinSpain:asmall-areastudy,Environ.Int.(November)(2016),doi:http://dx.doi.org/ 10.1016/j.envint.2016.11.009pii:S0160-4120(16)30755-3.

[6]EUROTHINE(TacklingHealthInequalitiesInEurope),EUROTHINEHomepageAvailable:,(2005).(Accessed05October 2018) http://www.irdes.fr/english/international-collaborations-and-european-projects/eurothine-tackling-health-inequalities-in-europe.html.

[7]M.Mircea,L.Ciancarella,G.Briganti,G.Calori,A.Cappelletti,I.Cionni,M.Costa,G.Cremona,M.D’Isidoro,S.Finardi,G.Pace, A.Piersanti,G.Righini,C.Silibello,L.Vitali,G.Zanini,AssessmentoftheAMS-MINNIsystemcapabilitiestopredictair qualityoverItalyforthecalendaryear2005,Atmos.Environ.84(2014)178–188.

[8]M.Mircea,G.Grigoras,M.D’Isidoro,G.Righini,M.Adani,G.Briganti,etal.,Impactofgridresolutiononaerosolpredictions: acasestudyoverItaly,AerosolAirQual.Res.16(2016)1253–1267,doi:http://dx.doi.org/10.4209/aaqr.2015.02.0058. [9]J.VandeKassteele,A.Stein,A.Dekkers,G.Velders,ExternaldriftkrigingofNOxconcentrationswithdispersionmodel

outputinareducedairqualitymonitoringnetwork,Environ.Ecol.Stat.16(3)(2009)321–339.

[10]R.Ignaccolo,S.Ghigo,S.Bande,Functionalzoningforairquality,Environ.Ecol.Stat.20(1)(2013)109–127.

[11]S.Ghigo,S.Bande,L.Ciancarella,M.Mircea,A.Piersanti,G.Righini,etal.,Mappingairpollutantsatmunicipalitylevelin ItalyandSpaininsupporttohealthimpactevaluations,AirQual.Atmos.Health11(1)(2017)69–82,doi:http://dx.doi.org/ 10.1007/s11869-017-0520-x.

[12]CORINE.CorineLandCover2006RasterData.Available: http://www.eea.europa.eu/data-and-maps/data/corine-land-cover-2006-raster-2(Accessed05October2018).

[13]C.Gariazzo,C.Silibello,S.Finardi,P.Radice,A.Piersanti,G.Calori,etal.,Agas/aerosolairpollutantsstudyovertheurban areaofRomeusingacomprehensivechemicaltransportmodel,Atmos.Environ.41(2007)7286–7303.

(10)

[14]P.Nafstad,L.L.Haheim,T.Wisloff,F.B.GramOftedal,L.Holme,etal.,Urbanairpollutionandmortalityinacohortof Norwegianmen,Environ.HealthPerspect.112(5)(2004)610–615,doi:http://dx.doi.org/10.1289/ehp.6684.

[15]C.Silibello,G.Calori,G.Brusasca,A.Giudici,E.Angelino,G.Fossati,etal.,ModellingofPM10concentrationsoverMilano urbanareausingtwoaerosolmodules,Environ.Model,Softw.23(2008)333–343.

[16]K.Thygesen,J.S.Alpert,A.S.Jaffe,M.L.Simoons,B.R.Chaitman,H.D.White,TheWritingGrouponbehalfoftheJointESC/ ACCF/AHA/WHFTaskForcefortheUniversalDeﬁnitionofMyocardialInfarction,Thirduniversaldeﬁnitionofmyocardial infarction,J.Am.Coll.Cardiol.60(16)(2012),doi:http://dx.doi.org/10.1016/j.jacc.2012.08.001(2012mISSN0735-1097).